An Introduction to Complex Systems Science and its Applications

Alexander F. Siegenfeld1,2∗ and Yaneer Bar-Yam2

1
Department of Physics, Massachusetts Institute of Technology, 77 Massachusetts Ave., Cambridge, MA and
2
New England Complex Systems Institute, 277 Broadway, Cambridge, MA

The standard assumptions that underlie many conceptual and quantitative frameworks do not
hold for many complex physical, biological, and social systems. Complex systems science clarifies
when and why such assumptions fail and provides alternative frameworks for understanding the
properties of complex systems. This review introduces some of the basic principles of complex
systems science, including complexity profiles, the tradeoff between efficiency and adaptability, the
arXiv:1912.05088v1 [physics.soc-ph] 11 Dec 2019

necessity of matching the complexity of systems to that of their environments, multi-scale analy-
sis, and evolutionary processes. Our focus is on the general properties of systems as opposed to
the modeling of specific dynamics; rather than provide a comprehensive review, we pedagogically
describe a conceptual and analytic approach for understanding and interacting with the complex
systems of our world. With the exception of a few footnotes, this paper assumes only a high school
mathematical and scientific background, so that it may be accessible to academics in all fields,
decision-makers in industry, government, and philanthropy, and anyone who is interested in systems
and society.

I. INTRODUCTION II. BASIC PRINCIPLES OF COMPLEX

SYSTEMS SCIENCE

How can we scientifically approach the study of com-
plex systems—physical, biological, and social? Empirical
studies, while useful, are by themselves insufficient, since
all experiments require a theoretical framework in which
they can be interpreted. While many such frameworks
exist for understanding particular components or aspects
of systems, the standard assumptions that underlie most
quantitative studies often do not hold for systems as a
whole, resulting in a mischaracterization of the causes
and consequences of large-scale behavior.
This paper provides an introduction to complex sys-
tems science, demonstrating a few of its applications and
its capacity to help us make more effective decisions in
the complex systems of our world. It focuses on the gen-
eral properties of complex systems, rather than on the
modeling of specific dynamics as in the (perhaps more
well-known) subfields of dynamical systems, agent-based
modeling, network science, and chaos theory. Section II
introduces key concepts, including complexity profiles,
the tradeoff between efficiency and adaptability, and the
necessity of matching the complexity of systems to that
of their environments. Section III considers the anal-
ysis of complex systems, attending to the oft-neglected
question of when standard assumptions do and—more
importantly—do not apply. Section IV discusses princi-
ples for effectively intervening in complex systems given
that their full descriptions are often beyond the limits of
human comprehension. Section V concludes.
∗ Corresponding author. Email: asiegenf@mit.edu
A. Why complex systems science?
Complex systems science considers systems with many
components. These systems could be physical, biolog-
ical, or social. Given this diversity of systems, it may
seem strange to study them all under one framework.
But while most scientific disciplines tend to focus on the
components themselves, complex systems science focuses
on how the components within a system are related to
one another [2]. For instance, while most academic disci-
plines would group the systems in fig. 1 by column, com-
plex systems science groups them by row. Given that the
principles of complex systems science are broadly appli-
cable, it should not be surprising that some have been in-
dependently discovered in particular domains. The pur-
pose of complex systems science is to provide a unifying
scientific framework, which allows for the generalization
(and therefore sharpening) of ideas and for the discovery
of new applications and connections.
A full description of all the details of even relatively
simple systems is impossible; therefore sound analyses
must seek the properties of systems that do not depend
on all of the details. That such properties exist is due
to universality, a phenomenon that will be discussed in
section III. An underlying insight that allows for the dis-
covery of such properties comes from a lesson of statisti-
cal physics, namely that while attempting to character-
ize the behavior of a particular state of a system (e.g. a
gas) may be entirely intractable, characterizing the set
of all possible states may not only be tractable but may
also provide us with a model of the relevant information
(e.g. the pressure, temperature, density, compressibility,
etc.). In other words, considering the space of possible
behaviors provides a powerful analytical lens that can be

Examples of Behaviors
Gas Pond Crowds
Random
Life
Canon Ball Infections Armies
Coherent
Snowflake Humans Corporations
Correlated
Physical Biological Social
FIG. 1. From [1]. Each column contains three examples of
systems consisting of the same components (from left to right:
molecules, cells, people) but with different relations between
them. Each row contains systems in which the relationship
between the components is the same. For random systems,
the behavior of each component is independent from the be-
havior of all other components. For coherent systems, all
components exhibit the same behavior; for example, the be-
havior (location, orientation, and velocity) of one part of the
cannonball completely determines the behavior of the other
parts. Correlated systems lie between these two extremes,
such that the behaviors of the system’s components do de-
pend on one another, but not so strongly that every com-
ponent acts in the same way; for example, the shape of one
part of a snowflake is correlated with but does not completely
determine the shape of the other parts. (Implicit in these de-
scriptions is the necessity of specifying the set of behaviors
under consideration, as discussed in section II B.)
applied not only to physical systems but also to biolog-
ical and social ones. And just as a system’s entropy, a
measure of its number of possible states, plays a pivotal
role in physics, a system’s complexity, a generalization of
entropy, plays a pivotal role in complex systems science.
B. What is complexity?
Simply put, the complexity of a behavior is equal to
the length of its description. The length of a description
of a particular system’s behavior depends on the number
of possible behaviors that system could exhibit. For ex-
ample, a light bulb that has two possible states—either
2
on or off—can be described by a single bit: 0 or 1. Two
bits can describe four different behaviors (00, 01, 10, or
11), three bits can describe eight behaviors, and so on.
Mathematically, we can write C = log2 N , where C is
complexity and N is the number of possible behaviors,1
but for our purposes here, it is sufficient to state that the
greater the number of possible behaviors, the greater the
complexity.
It is important to note that one must carefully define
the space of possible behaviors. For instance, if we are
interested in a light bulb already in a socket, the light
bulb has two possible behaviors, as above, but if we are
instead interested in the complexity of building a light
bulb, the space of possible behaviors might include all of
the ways in which its parts could be arranged. As another
example, consider programming a computer to correctly
answer a multiple-choice question with four choices. At
first glance, this task is very simple: since there are four
possible behaviors, only two bits are required. Nonethe-
less, we have the sense that programming a computer to
score perfectly on a multiple-choice test would be quite
difficult. This apparent paradox is resolved, however,
when we recognize that such a task is difficult only be-
cause we do not a priori know what questions will be on
the test, and thus the true task is to be able to correctly
answer any multiple-choice question. This task is quite
complex, given the large number of possible ways the
program could respond to a string of arbitrary multiple-
choice questions.
C. What is a complex system?
Consider a human, and then consider a gas containing
the very same molecules that are in the human but in no
particular arrangement. Which system is more complex?
The gas possesses a greater number of possible arrange-
ments of the molecules (i.e. has more entropy, or disor-
der), so it has the greater complexity. However, when
we think of a complex system, we think of the carefully
ordered arrangement of molecules in a human, not the
maximally disordered arrangement of a gas. Plants grow
by using energy from sunlight, decreasing the entropy of
their constituent molecules (although the net entropy of
the universe still increases). A city consists of physical
infrastructure intertwined with social, political, and eco-
nomic institutions and the people whose behaviors drive
those institutions. It therefore may be tempting to con-
clude that complex systems are those with reduced disor-
der. But the systems with the least disorder are those in
1 Technically, log2 N is actually an upper bound for the complex-
ity, since if some behaviors are more likely than others, the av-
erage length of the description can be reduced by using shorter
descriptions for the more common behaviors and longer descrip-
tions for the less common ones. (Lossless compression algorithms
rely on this logic.)

which all components exhibit the same behavior (coher-
ent systems in fig. 1), and such behavior is not intuitively
complex.
To resolve this paradox, we must consider that the
length of a system’s description depends on the level of
detail used to describe it. Thus, complexity depends on
scale. On a microscopic scale, the gas really is more com-
plex than the human: it is more difficult to describe the
positions and velocities of all of the molecules of the gas
than it is to do the same for all of the molecules of the
human. But at the scale of human perception, the range
of behaviors of a gas can be described by its tempera-
ture and pressure, while the behaviors of a human remain
quite complex at this scale. Entropy corresponds to the
amount of complexity at the smallest scale, but charac-
terizing a system requires understanding its complexity
across multiple scales. A system’s complexity profile is a
plot of the system’s complexity as a function of scale [2].
In the examples below, scale will be taken to be length,
but fundamentally, the scale of a behavior is equal to
the number of coordinated components of the system in-
volved,2 for which physical length is a proxy. A gas is
very simple at the scale of human perception, because at
this scale, only behaviors involving trillions of molecules
are relevant, and there are relatively few distinguishable
behaviors of a gas involving so many molecules.
As shown in fig. 2, random, coherent, and correlated
systems (see fig. 1) have qualitatively different complex-
ity profiles. Random systems have the most complexity
at the smallest scale (finest granularity/most detail), but
the amount of complexity rapidly drops off as the scale
is increased and the random behaviors of the individual
components are averaged out. A coherent system has
the same amount of complexity at small scales as it does
at larger scales because describing the overall behavior
of the system (e.g. the position and velocity of a canon
ball) also describes the behavior of all of the components
(e.g. the positions and velocities of all of the atoms).
Note that complexity tends to increase (or remain the
same) as the scale decreases, since looking at a system
in more detail (while still including the whole system in
the description) tends to yield more information.3 For
a correlated system, various behaviors occur at various
scales, and so the complexity gradually increases as one
2 Modeling each component of the system as a random variable
Xi and letting X = {X1 , ..., XN } denote the set of components
in the system, the complexity at scale s can be written as
X there to be complexity at larger scales, there must be be-
C(s) = I(A|X − A)
A⊂X s.t. |A|≥s
where I(A|B) denotes the multivariate mutual information be-
tween all of the random variables in set A conditioning on the
random variables in set B [3].
3 As a technical aside, constraints among three or more compo-
nents of a system (such as a parity constraint) can result in
negative complexity at certain scales [3], but such behavior is
beyond the scope of this introduction.
3
Random
complexity
Correlated
Coherent
scale
FIG. 2. Representative complexity profiles for random, cor-
related, and coherent systems.
examines the system in greater and greater detail. For
instance, from very far away, a human, being barely vis-
ible, has very little complexity. As the level of detail
is gradually increased, the description will first include
the overall position and velocity of the human, and then
the positions and velocities of each limb, followed by the
movement of hands, fingers, facial expressions, as well
as words that the human may be saying. Continuing to
greater levels of detail, the organs and then tissues and
patterns within the human brain become relevant, and
eventually so do the individual cells. At scales smaller
than that of a cell, complexity further increases as one
sees organelles (cellular substructures), followed by large
molecules such as proteins and DNA, and then eventu-
ally smaller molecules and individual atoms. At each
level, the length of the description grows longer. This
incredible multi-scale structure with gradually increas-
ing complexity is the defining characteristic of complex
systems.
D. Tradeoffs between complexity and scale
The intuition that complexity requires order is not un-
founded: it is no coincidence that the formation of com-
plex systems involves a decrease in entropy, i.e. a de-
crease in complexity at the smallest scale. In order for
haviors involving the coordination of many smaller scale
components. But this coordination suppresses complex-
ity at smaller scales because the behaviors of the smaller-
scale components are now limited by the interdependen-
cies between them. This tension can be made precise:
given a fixed set of components with a fixed set of poten-
tial individual behaviors, the area under the complexity
curve will be constant, regardless of the interdependen-
 Factory
complexity
Many types of goods but A corollary of the tradeoff between complexity and
only a few of each type
complexity
Many copies of a few types of goods largest possible scale. Thus, due to its low complexity,
scale
scale A very adaptable system, designed to handle all sorts
FIG. 3. The complexity profile of a factory that can produce
a large number of copies of a few types of goods, and the
complexity profile of a factory that can produce many types
of goods but not in large numbers.
cies (or lack thereof) between the components [3].4 Thus,
for any system, there is a fundamental tradeoff between
the number of behaviors a system can have and the scale
of those behaviors.
For instance, consider a factory consisting of many
workers [1]. The number of different types of goods that
the factory can produce at a given scale is a proxy for
the factory’s complexity at that scale, with the number
of copies of the same type of good that the factory can
produce in a given amount of time being a proxy for scale.
The fundamental tradeoff is evident in the fact that if the
factory wants to be able to churn out many copies of a
single good in a short amount of time, it will have to
coordinate all of its workers (perhaps having them work
on an assembly line), thereby reducing their individual
freedom to make many different kinds of goods. The fac-
tory’s production would then have low complexity but at
a large scale (e.g. churning out many identical Model-T
Fords—“Any customer can have a car painted any color
that he wants so long as it is black”). On the other hand,
if the factory’s employees work independently, they will
be able to create many different types of products, but
none at scale (fig. 3). Of course, a factory may change its
configuration of workers over time; the above analysis ap-
plies only to fixed configurations. Also, a factory may be
able to increase both the complexity and scale of its pro-
duction by adding new machinery or more workers; the
tradeoff between complexity and scale applies only when
4 If complexity profiles are formally defined as they are in the
previous footnotes, then the sum rule can be expressed as
X X
C(s) = H(Xi )
s i
where H(Xi ) is the unconditional entropy of Xi , which does not
depend on any of the correlations between the random variables.
4
considering the various collective behaviors of a fixed set
of components with a fixed set of individual behaviors.5
scale is the tradeoff between adaptability and efficiency.
Adaptability arises when there are many possible actions
happening in parallel that are mostly independent from
one another, i.e. when the system has high complexity.
Efficiency, on the other hand, arises when many parts of
a system are all working in concert, so that the system
can perform the task for which it was designed at the
a very efficient system will necessarily not be adaptable
to unforeseen variations within itself or its environment.
of shocks, will necessarily have to sacrifice some larger-
scale behaviors. The Soviets thought they could have
their cake and eat it, too: they originally believed that
their economy would outperform capitalist ones because
capitalist economies have so much waste related to multi-
ple businesses competing to do the same thing [4, Chap-
ter 16]. It would be far more efficient to have everyone
coordinated. But in creating such large-scale economic
structures, lower-scale complexity was sacrificed, result-
ing in a non-adaptive system. Of course, improperly reg-
ulated capitalist systems may also become maladaptive
due to, for instance, an excessive concentration of market
power, harmful feedback loops, or herd-like behaviors [5].
Due to the tradeoff between complexity and scale, any
mechanism that creates larger scale complexity—whether
market or government or otherwise—will necessarily re-
duce individual complexity. This is not to say that larger-
scale complexity is always harmful; it is often worth trad-
ing some individual-level freedoms for larger-scale coop-
eration. When, then, is complexity at a particular scale
desirable?
E. Why be complex?
A determination of when complexity is desirable is pro-
vided by the Law of Requisite Variety: To be effective, a
system must be at least as complex as the environmen-
tal behaviors to which it must differentially react. If a
system must be able to provide a different response to
each of 100 environmental possibilities and the system
has only 10 possible actions, the system will not be effec-
tive. At the very least the system would need 100 possible
actions, one for each scenario it could encounter. (The
above condition is necessary but of course not sufficient;
5 A subtle point to be made here is that introducing interactions
between two parts of a system may in some cases increase the set
of individual behaviors of each part, thereby increasing the total
area under the complexity profile. For example, if two people
enter into communication with each other, the communication
itself (e.g. speech) may now be a relevant behavior of each indi-
vidual person that was not there before.

complexity
larger scale scale behavior but lower complexity (e.g. the U.S. army
higher complexity
complexity
higher complexity at smaller scales livering vaccines is lower complexity but higher scale be-
complexity
higher complexity tives are suited for large-scale, lower complexity tasks,
at larger scales but tasks like case management require health systems
scale The Eurozone provides a potential illustration of a
FIG. 4. Schematic complexity profiles of militaries in con-
flict. Top: If two armies are operating with the same number
of possible behaviors but at different scales, the larger scale
one is favored. Middle: If two armies are operating at the
same scale but with different numbers of possible behaviors,
the higher-complexity one is favored. Bottom: If two armies
are operating at different scales and with different numbers
of possible behaviors, which one is favored depends on the
terrain (see text). Note that these profiles are simplified to
highlight the key concepts; actual militaries operate at multi-
ple scales. More generally, the top and middle graphs depict
conflicts in which one army has at least as much complexity
as the other at every scale.
a system with sufficiently many actions may still not take
the right actions in the right circumstances.)
Since complexity is defined only with respect to a par-
ticular scale, we can refine the Law of Requisite Variety:
To be effective, a system must match the complexity of
the environmental behaviors it is attempting to differen-
tially react to at all scales for which these behaviors oc-
cur [3]. To illustrate this multi-scale version of the Law
of Requisite Variety, we consider military conflict [6] (see
fig. 4). Here, one military can be considered as the sys-
tem, while the other military is part of the environment
with which the system must interact. For two militaries
of equal complexity, i.e. with the same number of behav-
iors, but with one military operating at a larger scale (e.g.
two very tightly controlled armies, but with one army
larger than the other), the larger-scale military will likely
win. For two militaries of equal scale but unequal com-
plexity (e.g. two equally sized and equally powered fleets,
but with one being more maneuverable than the other),
the higher-complexity military will likely win, since the
high-complexity military has an action for every action of
the lower-complexity military but not vice versa. When
5
a military with high complexity at a smaller scale (e.g.
a guerrilla force) conflicts with a military with larger-
in Vietnam or the Soviet army in Afghanistan), the ter-
rain, which constrains the scale of the conflict, plays an
important role. In an open field, or in open waters, the
military that has more complexity at the larger scales is
favored, while in the jungle or in the mountains, higher
complexity at smaller scales is favored.
As another example, healthcare involves both small-
scale tasks with high overall complexity such as case man-
agement, as well as large-scale, lower complexity tasks,
such as manufacturing and delivering vaccines [7]. (De-
cause the same actions will be performed for nearly ev-
eryone.) Large-scale top-down organizations and initia-
with a high degree of small-scale (i.e. local) complexity.
multi-scale complexity mismatch. Fiscal policy is made
predominantly at the scale of individual countries and
thus has a higher complexity at the country scale but
relatively little complexity at the scale of the entire Eu-
rozone, while monetary policy is made at the scale of
the entire Eurozone and thus has some complexity at
the scale of the Eurozone but lacks the ability to vary
(i.e. lacks complexity) at the scale of individual coun-
tries. Many have argued that economic difficulties within
the Eurozone have arisen because this mismatch has pre-
cluded effective interactions between fiscal and monetary
policy [8–12].
In none of the above examples have the complexity
profiles been precisely calculated, nor has scale been pre-
cisely defined. Instead, proxies for scale are used and
estimated comparisons of complexity made. Such an
approach cannot yield precise results (indeed, no ap-
proach can, given the complexity a full description of such
systems would require), but additional precision is not
needed when even the approximate analysis reveals large
mismatches in complexity.6 (To remedy the diagnosed
mismatches, more detailed analyses may be required.)
Complexity (at any scale) is neither intrinsically good
nor bad. Instead, problems arise from mismatches be-
tween the complexity of a task to be performed and
the system performing that task. (It is important
to note that the system in one scenario may be the
task/environment in another; for instance, the same com-
6 Just as this analysis of the space of possible behaviors can be
used even in the face of uncertainty regarding a system’s precise
mechanisms and outcomes, physicists can use the property of en-
tropy (sometimes considering how quantities related to entropy
change across scale) to classify phase transitions even when they
cannot, from first principles, determine precise quantities such as
the amount of heat generated by a phase transition or the tem-
perature at which it occurs (such quantities must be determined
empirically).
 6

plexity that helps a system interact with its environment etc.) may still hinder progress on problems that do not
may prevent its effective management by other systems.) fall neatly within a discipline or sub-discipline [14–19].
Incidentally, human emotions appear to reflect this prin- The above examples provide an illustration of the prin-
ciple: we are bored when our environment is too simple ciple that in order for a system to differentially react to
and overwhelmed when it is too complex [13]. a certain set of behaviors of its environment, not only
must the system as a whole have at least as much com-
plexity at all scales as this set of environmental behav-
F. Subdivided systems iors (as described in section II E), but also each subset
of the system must have at least as much complexity at
all scales as the environmental behaviors corresponding
Even if the complexity of the system matches that of
to that subset. A good rule of thumb for applying this
its environment at the appropriate scales, there is still
principle is that decisions concerning independent parts
the possibility of a complexity mismatch. Consider two
or aspects of a system should be able to be made indepen-
pairs of friends—four people total, each of whom can lift
dently, while decisions concerning dependent parts of the
100 pounds—and consider two 200-pound couches that
system should be made dependently. It follows that the
need to be moved. Furthermore, assume that each per-
organizations that make such decisions should be subdi-
son is able to coordinate with her friend but not with ei-
vided accordingly, so that their subdivisions match the
ther of the other two people. Overall then, the system of
natural divisions in the systems with which they inter-
people has sufficient complexity at the appropriate scales
act.7
to move both couches since each pair of friends can lift
one of the 200-pound couches. However, were one person
from each pair of friends to be assigned to each couch, G. Hierarchies
they would not be able to lift the couches because the two
people lifting each couch would not belong to the same
A common way of organizing systems is through con-
pair of friends and thus would not be able to coordinate
trol hierarchies. In an idealized hierarchy, there are no
their actions. The problem here is that while the pairs
lateral connections: any decision that involves multiple
of friends possess enough overall complexity at the right
components of the hierarchy must pass through a com-
scales to lift the couches, the subdivision within the sys-
mon node under whose jurisdiction these components all
tem of friends is not matched to the natural subdivision
(directly or indirectly) lie. The complexity profile of such
within the system of couches. The mismatch in complex-
a hierarchy depends on the rigidity of the control struc-
ity can be seen if we focus our attention on just a single
ture (fig. 5). At one extreme, every decision, no matter
couch: while the couch requires coordinated action at the
how large or small, is made by those at the top of the
scale of 200 pounds, the two people lifting it are capable
hierarchy. This hierarchy has the same amount of com-
only of two independent actions, each at the scale of 100
plexity across all its scales: namely the complexity of
pounds.
whatever decisions are being made at the top. At the
The way in which academic departments are organized
other extreme, there is no communication within the hi-
provides a more realistic example of the potential of sub-
erarchy, and every individual acts independently. This
division mismatch. Academia has multiple levels of sub-
hierarchy has very little complexity beyond the individ-
division (departments, subfields, etc.) in order to or-
ual level. Between these two extremes is a typical hier-
ganize knowledge and coordinate people, resulting in a
archy, in which different decisions are made at different
high overall degree of complexity across multiple scales,
levels.
where scale could refer to either the number of coordi-
No type of hierarchy is inherently better than any
nated people or the amount of coordinated knowledge,
other. For a particular environment, the best hierarchy is
depending on which aspect of the academic system is un-
one whose complexity profile matches the tasks that it is
der consideration. Similarly, there are multiple levels of
trying to perform. A tightly controlled, or top-heavy, hi-
natural subdivision in the set of problems that academia
erarchy is not well suited to environments in which there
can potentially address, with each subdivision of prob-
is a lot of variation in the systems with which the lower
lems requiring particular types of coordinated knowledge
levels of the hierarchy must interact; neither is a very
and effort in order to be solved. Academia’s complex-
loosely controlled hierarchy well suited to environments
ity across multiple scales allows it to effectively work
that require large-scale coordinated action. For exam-
on many of these problems. However, there may exist
ple, centralizing too much power within the U.S. gov-
problems that academia, despite having sufficient overall
ernance system at the federal (as opposed to the local
multi-scale complexity, is nonetheless unable to solve be-
cause the subdivisions within the problem do not match
the subdivisions within academia. The increase in in- 7 The subdivisions present in the human brain and the analysis of
terdisciplinary centers and initiatives over the past few subdivisions in neural networks more generally [2, Chapters 2.4-
decades suggests the perception of such a mismatch; how- 2.5] demonstrate how systems that are subdivided so as to match
ever, the structure of the academic system as a whole (in- the natural subdivisions in their environments outperform those
cluding how students are trained, the publishing process, with more internal connectivity.

Lower levels given
some autonomy
complexity
Tightly controlled a stark example: the allocation of resources and labor
scale allocate resources are made without any individual mak-
FIG. 5. Complexity profiles of two hierarchies, each with
the same number of people. Here, the scale is the number
of coordinated man-hours. In one hierarchy, all decisions,
regardless of the scale, are made by a single person, while in
the other, different decisions are made at various levels of the
hierarchy.
or state) level would not allow for sufficient smaller-scale
complexity to match the variation among locales; too de-
centralized a system would not allow for sufficient larger-
scale complexity to engage with problems that require
nationally coordinated responses.8 Assigning decisions
to higher levels in hierarchies allows for more efficiency
and scale, but such decisions result in less adaptabil-
ity because when they are incorrect, they affect more
of the system and—as larger-scale changes tend to re-
quire longer time-scales to enact—are more difficult to
roll back.
It is important to distinguish between the complexity
of a hierarchy and the complexity of the decisions that
the people within the hierarchy are capable of making.
For instance, one could design a tightly controlled hierar-
chy that could take a large number of large-scale actions
(i.e. high complexity at its highest scale), but, since the
decision-making abilities of even the most capable hu-
mans are of finite complexity, the individuals at the top
may lack the complexity to correctly choose from among
these actions. This brings us to an important limitation
8 We can also consider not just the overall complexity profile
of governance, but how well subdivisions in government match
those within its environment (section II F). Metropolitan areas
are in some ways more similar to one another than they are to
the rural areas of their respective states. So while dividing the
U.S. into 50 states provides substantial lower-scale governmental
complexity, this complexity is not necessarily well matched to
natural urban-rural divides. To the extent that such a mismatch
exists, there may be issues currently handled at the state level
that would be better handled at the local level, thereby allow-
ing for different policies in urban and rural areas (and likewise,
perhaps some of the powers that some argue should be devolved
from the federal to the state level should in fact be devolved to
the local level).
7
of hierarchies: the complexity of the decisions concerning
the largest-scale behaviors of a hierarchy—the behaviors
involving the entire organization—is limited by the com-
plexity of the group of people at the top [1]. Thus, a hier-
archy will necessarily fail when the complexity of match-
ing its largest-scale behaviors to those of the system it
is attempting to regulate9 is higher than the complex-
ity of decision-making achievable by any individual or
committee. The failure of command economies provides
is too complex a problem for any one person or group
of people to understand. Markets allocate resources via
a more networked system: decisions regarding how to
ing them, just as decisions are made in the human brain
without any neuron making them. (Whether or not these
market allocations are desirable depends in part on the
way in which the market is structured and regulated.)
We began by considering idealized hierarchies with
only vertical connections, but lateral connections pro-
vide another mechanism for enabling larger-scale behav-
iors. For instance, cities can interact with one another
(rather than interacting only with their state and na-
tional governments) in order to copy good policies and
learn from each other’s mistakes. Through these sorts of
evolutionary processes (described further in section IV),
large-scale decisions (large-scale because policies may be
copied by multiple cities) that are more complex than any
individual component can be made. Such lateral connec-
tions can exist within a hierarchical framework in which
the top of the hierarchy (in this example, the national
government) maintains significant control, or they can
exist outside of a hierarchical structure, as in the human
brain. Furthermore, these lateral connections can vary
in strength; overly strong connections lead to herd-like
behavior with insufficient smaller-scale variation such as
groupthink [20–22] (no system is exempt from the trade-
off described in section II D), while overly weak connec-
tions result in mostly independent behavior with little
coordination.
III. ANALYZING COMPLEX SYSTEMS
The previous section has examined some of the general
properties of systems with many components. But how
9 Note that the complexity of the task of deciding which behaviors
of a system should correspond to which behaviors of its environ-
ment is generally much greater than the complexity of either the
system or the environment alone: for example, if both the sys-
tem and environment have 10 possible behaviors, the system has
enough complexity to match the environment, but properly de-
ciding which behaviors of the system should correspond to which
environmental conditions requires correctly choosing one option
out of a space of 3,628,800 (10 factorial) possibilities. The space
of possible behaviors of the system and its environment may be
much smaller than the space of possible decisions concerning the
management of the system’s actions in its environment.

complexity
s0 scale captured by the average behavior of the system, i.e. so
FIG. 6. A complexity profile of a system with a separation
of scales. A separation of scales implies that the behaviors
occurring below a certain scale (s0 in the above figure) are, at
larger scales, mostly independent from one another and that
therefore, at these larger scales, only the average effects of
these behaviors are relevant.
do we study particular systems? How do we analyze data
from complex systems, and how do we choose which data
to analyze?
A. How do we understand any system?
In a sense, it is surprising that we can understand any
macroscopic system at all, as even a very simple mechan-
ical system has trillions upon trillions of molecules. We
are able to understand such systems because they possess
a separation of scales [23], meaning that the macroscopic
behavior we are interested in occurs at a far larger scale
than the behavior of the individual molecules, with not
much behavior occurring in between these two scales (see
fig. 6). This separation allows us to treat the macroscopic
and microscopic behavior separately: for mechanical sys-
tems, we treat the macroscopic behavior explicitly with
Newtonian mechanics, and the microscopic behavior is
considered in aggregate using thermodynamics.
More generally, the approach described above is an ex-
ample of a mean-field theory [24], in which the average be-
haviors of a system’s components are explicitly modeled
and the deviations of the individual components from
this average are treated as statistically independent ran-
dom fluctuations. This approach works very well for sys-
tems such as computers, cars, airplanes, and buildings, in
which the motions of individual molecules are, apart from
some mostly uncorrelated fluctuations, well described by
the motion of the piece of material to which they belong.
Such an approach is also often employed in analyses of bi-
ological, social, and economic systems; these assumptions
work well in many cases, but as we will see, they are not
always appropriate for complex systems. It is important,
therefore, to determine under what conditions mean-field
8
theory holds.
B. When mean-field theory breaks down
The systems for which mean-field theory applies ex-
hibit large-scale behaviors that are the average of the
behaviors of their components. They must possess a sep-
aration of scales, which arises when the statistical fluc-
tuations of their components are sufficiently independent
from one another above a certain scale. Mean-field the-
ory may hold even in the presence of strong interactions,
so long as the effect of those strong interactions can be
long as each component of the system can be modeled
as if it were interacting with the average of the system
(i.e. with the mean field). For example, the large-scale
motion of solids is well described by mean-field theory,
even though the molecules in a solid interact with one
another quite strongly, because the main effect of these
interactions is to keep each molecule at a certain dis-
tance and orientation from the average location (center
of mass) of the solid. Likewise, under some (but certainly
not all) conditions, economic markets can be effectively
described by modeling each market actor as interacting
with the aggregate forces of supply and demand rather
than with other individual market actors.
However, when there are sufficiently strong correla-
tions between the components of the system, i.e. when
the interactions between a component of the system and
a specific set of other components (as opposed to its gen-
eral interaction with the rest of the system) cannot be ne-
glected, mean-field theory will break down.10 These sys-
tems will instead exhibit large-scale behaviors that arise
not solely from the properties of individual components
but also from the relationships between components. For
example, while the behavior of a muscle can be roughly
understood from the behavior of an individual muscle
cell, the behavior of the human brain is fundamentally
different from that of individual neurons, because cog-
nitive behaviors are determined largely by variations in
the synapses between neurons. Similarly, the complex
ecological behaviors of a forest cannot be determined by
the behaviors of its constituent organisms in isolation.
Because their small-scale random occurrences are not
statistically independent, complex systems often exhibit
large-scale fluctuations not predicted by mean-field the-
ory, such as forest fires, viral content on social media, and
crashes in economic markets. Sometimes, these large-
scale fluctuations are adaptive: they enable a system to
collectively respond to small inputs [25]. For instance,
humans respond strongly to minor disturbances in the
10 Incidentally, the failure of mean-field theory to describe cer-
tain physical phase transitions led physicists to develop a new,
multi-scale approach (the renormalization group), a foundation
of much of complex systems science.

1.2
1.0
probability density
0.8 Thin-tailed fall is astronomically low (literally so: there is about a
0.6
0.4
0.2
Fat-tailed ders might fall down together. Other examples include
0 2 4 6
scale in the possibility of global market crashes and the inter-
FIG. 7. A normal distribution (thin-tailed) and a distribution
with a power-law decay (fat-tailed). The fat-tailed distribu-
tion may appear more stable, due to the lower probability of
small-scale fluctuations and the fact that samples from the
distribution may not contain any extreme events. However,
sooner or later, a fat-tailed distribution will produce an ex-
treme event, while one could wait thousands of lifetimes of the
universe before a normal distribution produces a similarly ex-
treme event. Note that the axes of this graph are truncated;
the illustrated fat-tailed distribution can, with small but non-
negligible probability (0.04%), produce events with a scale of
one million or more.
density of air, such as the sound of their own names.
However, these large-scale fluctuations sometimes pose
systemic risks.
C. Fat-tailed distributions and systemic risk
When the components of a system are independent
from one another above a certain scale, then at much
larger scales, the magnitude of the fluctuations of the sys-
tem follow a normal distribution (bell curve),11 for which
the mean and standard deviation are well-defined and for
which events many standard deviations above the mean
are astronomically improbable. Interdependencies, how-
ever, can lead to a distribution of fluctuations in which
the probability of an extreme event, while still small, is
not astronomically so. Such distributions are character-
ized as fat-tailed —see fig. 7. For example, while human
height follows a thin-tailed distribution, with no record of
anyone over twice as tall as the average human, human
wealth—due to the complex economic interactions be-
tween individuals—follows a fat-tailed distribution, with
multiple individuals deviating from the average by fac-
tors of more than one million [26].
One danger of interdependencies is that in the short
term, they may make systems appear more stable by
11 This follows from the central limit theorem.
9
reducing the extent of small-scale fluctuations. As a
thought experiment, imagine 100 ladders, each with a
1/10 probability of falling. If the ladders are indepen-
dent from one another, the probability that all of them
1020 times higher chance of randomly selecting a particu-
lar atom out of all of the atoms in the known universe). If
we tie all the ladders together, we will have made them
safer, in the sense that the probability of any individ-
ual ladder falling will be much smaller, but we will have
also created a non-negligible chance that all of the lad-
the interconnectedness of our financial systems resulting
8
connectedness of travel routes increasing the probability
of pandemics [27] such as the Spanish flu. More gener-
ally, when the underlying probability distributions have
fat tails, standard statistical methods often break down,
leading to potentially severe underestimates of the prob-
abilities of extreme events [28].12
D. Understanding complex systems
Because it is usually easier to collect data regarding
components of a system than it is to collect data regard-
ing interactions between components, empirical studies
often fail to capture the information relevant to complex
systems, since complex large-scale behaviors critically de-
pend on such interactions. Furthermore, as discussed in
section III C, data analysis can severely underestimate
the probability of extreme events (tail risk). Finally, em-
pirical analyses often assume linearity, i.e. they assume
that the total impact of a set of factors is equal to the sum
of the impacts of each individual factor, an assumption
that often breaks down for complex systems.
How can we understand the systems for which these
standard approaches do not apply? Our understanding
of all systems with many components depends on uni-
versality [29], i.e. the existence of large-scale behaviors
that do not depend on the microscopic details. The stan-
dard approaches are predicated on the assumption of suf-
ficient independence between components, which allows
large-scale behaviors to be determined without a full ac-
counting of the system’s details via mean-field theory.13
But mean-field theory is just one example of universality.
12
For instance, the mean and variance of a fat-tailed distribution
may not exist, and even if they do, they may not be able to
be reliably estimated from a finite sample due to the likelihood
of a single data point or lack thereof substantially skewing the
estimate.
13 Formally, the statistical fluctuations of the components must,
above a certain scale, be sufficiently independent so as to satisfy
the assumptions of the central limit theorem. The central limit
theorem is the manifestation of universality that explains the
ubiquity of normal distributions.
 10

Information
FIG

all the details

dig
ar
Most important fac
sys
wit
large scale behavior sca

Scale
FIG. 8. A figure from Lim et al.’s paper on ethnic vio- lut
lence [30]. The sites where their model predicts a potential
for ethnic violence are shown in red in panels C and D, with FIG. 3: 9.The
FIG. complexity profile
A representative is the
complexity amount
profile of information
of a complex sys- ert
confirmed reports of ethnic violence depicted by the yellow thattem
is required to information
[23]. (Here, describe a system as a function
is synonymous of the scale
with complexity, fur
dots in panel D. as complexity Typically,
of description. is a measurelarger
of the amount of information
scales require fewernec-
details lar
and therefore smaller amounts of information. The scale.)
essary to convey a system’s behavior at a particular most im- mu
Understanding all the details (i.e. all of the small-scale be-
portant information about a system for informing action on
haviors) is impossible and unnecessary; the most important the
thatinformation
system is is thecontained
behaviorin at
thethe largest behaviors.
large-scale scale. How- inf
Sound is another example: all materials, regardless of
their composition, allow for the propagation of sound ever, for systems for which mean-field theory does not apply, pro
waves. Sound behaves so similarly in all materials be- characterizing these behaviors may be difficult.
siz
cause at the length scales relevant to sound waves, which at the finest scales is finite because of quantum uncer- tha
are far larger than the sizes of individual atoms and
tainty and is equal to a universal constant, 1/kB ln(2),
molecules, the effect of the microscopic parameters is
where ethnic violence has the potential to occur and ap-
merely to set the speed of the sound. Note that sound times the entropy for a system in equilibrium, where kB
14 scr
is plied their model
Boltzmann’s to India
constant and to what was Yugoslavia.
[15]. (F
waves cannot be understood as a property of the average
Ethnic violence has many causes, but rather than focus-
behavior—in this case, average density—of a material, A single real number has infinite possibilities in all of
ing on specific, culturally dependent mechanisms or on
les
since it is precisely the systematic correlations in the de- the infinite digits of its representation. Therefore it has the
the average properties of regions, such as demographic
viations from that average that give rise to sound. Nor is the ability to represent an infinite amount of informa- of
or economic statistics, the authors considered the multi-
sound best understood by focusing on the small-scale de-
tion. This
scale would
patterns in seem to indicate
how ethnic thatgeographically
groups were we could use a we
tails of atomic motion: scientists understood sound even
distributed
single real (fig. 8).toThey
number found that
represent a ethnic violence
system. did
For example, ext
before they learned what atoms are. The key to under-
a not occur
number when
that the ethnic the
represents groups were either
density of a well mixed
liquid has in- of
standing sound waves is to recognize that they have a
or well separated but rather occurred only when ethnic
multi-scale structure—with larger-scale fluctuations cor- finite information, but we know from phase transitions
groups separated into geographic patches,15 with the vio-
responding to lower frequencies and smaller-scale fluctua- that this single number isn’t enough. Why doesn’t this ph
lence being mostly likely to occur for geographic patches
tions corresponding to higher frequencies—and to model work? The problem is16that the way the information is or- sys
of a particular size. Although not explicitly included
them accordingly. ganized
in theinanalysis,
scale inspecific
the real number
details does not
of a region are correspond
relevant Re
Lim et al. apply this approach to studying ethnic vi-
olence [30]. They built a predictive model to analyze
to the way
insofar as it does
they are in the
either asystem. A
cause or an real
effect number
(or both) can mo
of the patch
represent size.17
the position of a point along one dimension. in
Let’s say we start by knowing where the object is to a ter
14 For Quantum Electrodynamics (the theory of how light and elec-
resolution of 1 unit of length. Increasing the resolution of
trons interact), we still do not know the microscopic details. Yet
by 15
a factor of two means we can distinguish which of the
This separation falls into the same universality class as the sep- act
we can still make predictions accurate to ten decimal places be- two possible segments
aration of oil and water.that are 1/2 units of length it is of
cause, as can be shown with renormalization group theory, the in. 16Communicating
This analysis implies this
that ethnic violence can
information be prevented
requires by
a binary is
only effect of these microscopic details at the scales at which we the use of well-placed political boundaries, as in Switzerland [31].
can make measurements is to set the electron mass and charge— variable. For each
17 For instance, 2 fold
animosity increase
between in resolution
two ethnic wenothave
groups, though ou
quantities that, like the speed of sound in any particular material, 2 additional possibilities
explicitly considered, may betoa specify. The
cause as well as a number
consequenceofofbits of
can be measured (but not predicted).
is thethe geographic segregation [32].
logarithm (base 2) of the scale of resolution. How- its
ever, for a liquid at its critical point the number of bits
increases di↵erently with increasing resolution. As res- us
olution increases we have to describe the fluctuations of ab
density. The growth in the number of bits is more rapid tio
than one bit per factor of two in resolution (see Fig. 4). we

Understanding all of the details of any complex sys-
tem is impossible, just as it is for most systems with a
separation of scales; there is just too much complexity at
the smallest scale. But unlike the behaviors of systems
with a separation of scales, the important large-scale be-
haviors of complex systems are not simply the average
of their small-scale behaviors. The interdependencies at
multiple scales make it unclear exactly how the small-
scale behaviors give rise to larger-scale ones. However,
even for complex systems, there is much less complexity
at the larger scales than there is at the smaller scales,
and so there will always be large-scale behaviors that do
not depend on most of the system’s details (see fig. 9). In
general, the key to understanding the large-scale behav-
ior of a system is to find the appropriate mathematical
(or conceptual) description, which for complex systems
is not a simple average nor a full account of all the de-
tails. Additional examples of this multi-scale approach
have been reviewed elsewhere [23].
IV. COMPLEX SYSTEMS AND UNCERTAINTY
Although the principles discussed throughout sec-
tions II and III help us recognize the fundamental prop-
erties and limitations of systems, our understanding of
most complex systems will inevitably be imperfect. And
regardless of how well-considered a plan is, a truly com-
plex system will present elements that were not consid-
ered ahead of time.18 Given the absence of perfect knowl-
edge, how can the success of systems we design or are
part of be assured? While the success of many systems
rests on the assumption that good decisions will be made,
some systems can perform well in spite of the fallibility
of the decision-makers (whether due to corruption, sub-
conscious bias, or the fundamental limitations of human
minds). The study of complex systems approaches this
observation scientifically by (implicitly or explicitly) con-
sidering the decision-makers themselves as part of the
system and of limited complexity/decision-making abil-
ity. The question thus becomes: how do we design sys-
tems that exceed the complexity of the decision-makers
within them?
18 It should also be noted that in a functional system with a high
degree of complexity, the potential positive impact of a change is
generally much smaller than its potential negative impact. For
example, a change to the wiring in a computer is unlikely to
dramatically improve the computer’s performance, but it could
cause the computer to crash. Airplanes are another example.
This phenomenon is a consequence of the fact that, by definition,
a high degree of complexity implies that there are many system
configurations that will not work for every one configuration that
will.
11
A. Evolutionary processes
A key observation is that while uncertainty makes
most systems weaker, some systems benefit from uncer-
tainty and variability [33–36]. The common character-
istic of these systems is their embodiment of some sort
of evolutionary process, i.e. a process in which success-
ful changes are copied (and further modified) while un-
successful changes are not. The classic evolutionary pro-
cesses are biological: due to variability introduced by ran-
dom mutations, organisms with the complexity and scale
of humans evolved from single-celled organisms. Further-
more, humans themselves have the property of benefiting
from exposure to random shocks (provided the shocks
are not too strong). Immune system performance is im-
proved by early exposure to non-lethal pathogens [37, 38];
muscles and bones are strengthened by micro-tears and
micro-fractures, respectively; we learn by exposure to
new information and problem-solving; and our psycholo-
gies are strengthened by exposure to adversity, provided
the adversity is not too severe [39, 40].
Competitive market economies provide another exam-
ple of how systems can thrive on uncertainty. Due to
our ignorance of which will succeed, many potential in-
novations and businesses must be created and improved
upon in parallel, the successful ones expanding and the
unsuccessful ones failing. The successful among these can
then be improved upon in the same manner—with many
approaches being applied at once—and so on. (How-
ever, without effectively regulated multi-scale coopera-
tive frameworks—see section IV B—large-scale parts of
the economic system may optimize for the wrong goals,
settling into harmful societal equilibria [41].)
Likewise, the internal processes of large organizations
may follow an evolutionary pattern in which small parts
of the organization can fail and thus be improved upon;
without such flexibility, the entire organization may fail
at once in the face of a changing external environment.
In some cases the failure of the entire organization makes
room for more effective organizations to take its place
(assuming the economy is sufficiently decentralized and
competitive so that the organization in question is not
“too big to fail”). The collapse of government is generally
not one of those cases, however [42], so it is especially
important that governance systems possess the flexibility
to internally benefit from randomness and uncertainty.
Perhaps counterintuitively, not allowing small failures to
occur may weaken systems in the long run by halting
evolutionary processes and by creating interdependencies
that lead to systemic risk (section III C).
In order to thrive in uncertainty and exceed the com-
plexity of individual decision-making, systems can incor-
porate evolutionary processes so that they, even if very
limited at first, will naturally improve over time. The
first step is to allow for enough variation in the system,
so that the system can explore the space of possibilities.
Since a large amount of variation means a lot of complex-
ity and complexity trades off with scale, such variation
competition between sports is what gives rise to the collaboration between teams, the
competition between teams gives rise to collaboration between players. Conversely
cooperation at each level enables competition at the higher level of organization. At the
same time competition and collaboration are antagonistic when they take place at the
same of organization unless they are made `orthogonal’ referring to different times or 12
types of behavior.

Competition compete with each other, and they too may be able to
between sports increase the effectiveness of their internal cooperation by
Team collaboration Competition between sports
enables the sport for fan attention and money introducing some healthy competition among their mem-
to exist and compete increases team collaboration bers (fig. 10 provides an example). If these members are
Collaboration Competition themselves groups, the process of competition begetting
between teams between teams
Competition between teams
cooperation that begets more competition can continue
Collaboration of
players enables causes selection of teams to even smaller scales. This process can work in reverse
teams to compete with collaborating players
as well: in order to compete more effectively, individuals
Collaboration
between players
Competition may cooperate with each other to form groups, which in
between players
turn may cooperate to form even larger groups, and so
Figure 3.
on. Thus, a complex network of cooperation and compe-
FIG.
The key 10.
point An
is thatillustration
competition andfrom [4, Chapter
cooperation 7] showing
always occur together at the
different
interplay betweenThis cooperation and competition
tition among groups of various sizes (scales) can naturally
levels of organization topic might be so intuitive to someinofthe
you context
that you are
of sportswhy
wondering teams and leagues.
I am talking about this at all. Surprisingly, it has not been clear to many
evolve.
in the context of scientific dialog about evolution. Even if understood intuitively in In order for it to promote effective group cooperation,
sports, and even if some scientists understand this intuitively as well, it is important to competition must be properly structured. A soccer team
state clearly these basic relationships. in which the players compete with their own team mem-
must occur at smaller scales (in both space and time).
For does
How example,
this help usin make
the effective
case of governance,
teams? The answer is,enabling
self-evident.each
Effective
bers to score goals will not be effective, but one in which
teams
city form naturally when there
to experiment is a process of evolutionary
independently allows forselection
manyofplans teams that the players compete for the title of the most fit may
perform well in competition. This may be a useful lesson for those who try hard to
to be tried out in parallel and to be iterated upon. The be. The framework in which competition occurs must
compel player behavior in one way or another. While I do not want to say that teaching be structured so that the competitors are incentivized
isopposite
not important.strategy
Still, it is would
the role ofbe to enact
competition itselfone national
to teach plan, I
about cooperation.
the effects
would of which
also say, that evolutionwill not
teaches be able about
us something to be the comparatively
proper place of rewards to take actions that are net good for the group; other-
for effective competition. The main reward is simply the right to stay together. This,
evaluated. wise a kind of tragedy-of-the-commons situation occurs.
after all, is what survival, survival of a collective, is all about. The potential for competition to go awry highlights the
The second step is to allow for a means of communica-
tion between various parts of the system so that success- importance of having a multi-scale structure with com-
ful choices are adopted elsewhere and built upon (e.g. petition occurring on multiple levels, rather than having
cities copying the successful practices of other cities). everyone in the system compete with everyone else. With
Plans will always have unintended consequences; the key the multi-scale structure, groups with unhealthy evolu-
is to allow unintended consequences to work for rather tionary dynamics are selected against, while groups with
than against the system as a whole. The desire for di- a healthy mix of competition and cooperation that bene-
rect control must often be relinquished in order to allow fits the entire group are selected for.20 Market economic
complexity to autonomously increase over time.19 systems are successful not because free markets produce
optimal outcomes (real-world markets often sharply de-
viate from the assumptions of free-market models, and
B. Multi-scale evolutionary processes externalities abound) but rather because, at their best,
appropriately regulated market systems allow for multi-
scale evolutionary processes to naturally arise, resulting
Successful evolutionary processes generally do not con- in innovations and complexity far beyond what anyone
sist of unbridled competition but rather contain both could have imagined, let alone designed.
competition and cooperation, each occurring at multiple
scales [43]. For example, cells cooperate within multicel-
lular organisms in order to more effectively compete with
other organisms, and organisms cooperate both within V. SUMMARY
and between species in order to more effectively compete
against other species. Competition at larger scales nat- Systems with many components often exhibit emergent
urally breeds cooperation at smaller scales, because in behaviors, i.e. behaviors that arise from the relationships
order for a group to effectively compete against another between the components rather than from the compo-
group (large-scale competition), there must be coopera- nents themselves. However, an early insight of statistical
tion within the group. Cooperation can also breed com- physics is that in spite of the impossibility of describing
petition since sometimes the best way for the group to the details of trillions of molecules, the macroscopic prop-
achieve its shared goals is to facilitate some healthy com- erties of the molecules can be well understood by analyz-
petition among its subgroups. Those subgroups must fos- ing their space of possible behaviors, rather than their
ter cooperation within themselves in order to effectively

20 There is evidence that the geographic nature of evolution—in

19 Systems can explicitly design only systems of lesser complexity which organisms evolve in somewhat separated environments and
since an explicit design is itself a behavior of the first system. mean-field theory does not apply—has resulted in precisely this
However, systems that evolve over time can become more com- multi-scale structure and has therefore allowed for the evolution
plex than their designers. of genuine (e.g. not reciprocal) altruistic behavior [44, 45].

specific configuration and motions. While many macro-
scopic properties can be described in terms of the average
behaviors of the molecules, the macroscopic properties of
certain physical phenomena, such as phase transitions,
cannot be understood by averaging over system com-
ponents; accordingly, physicists were forced to develop
new, multi-scale methods. Likewise, while standard sta-
tistical methods—which infer the average properties of a
system’s many components—can succesfully model some
biological and social systems, they fail for others, some-
times spectacularly so.
Taking a systemic view by considering the space of pos-
sible behaviors can yield insights that cannot be gleaned
by considering only the proximate causes and effects of
particular problems or crises. A system’s complexity—
which depends on its number of distinct potential be-
haviors (i.e. on the space of possibilities)—is a start-
ing point from which to get a handle on its large-scale
properties, in the same way that entropy is the start-
ing point for statistical physics. Because the number of
distinct behaviors of a system depends on the level of de-
tail (behaviors that appear the same at lower resolution
may be distinct at higher resolution), complexity depends
on scale. Interdependencies between components reduce
complexity at smaller scales by restricting the freedom
of individual components while creating complexity at
larger scales by enabling behaviors that involve multiple
components working together. Thus, for systems that
consist of the same components, there is a fundamental
tradeoff between the number of behaviors at smaller and
larger scales. This tradeoff among scales is related to the
tradeoff between a system’s adaptability, which depends
on the variety of different responses it has to small-scale
disturbances, and its efficiency, which depends on its op-
erating scale. There is no ideal scale at which a system
should possess complexity; rather, the most effective sys-
[1] Bar-Yam, Y. Complexity rising: From human beings to
human civilization, a complexity profile. In Encyclopedia
of Life Support Systems (EOLSS UNESCO Publishers,
Oxford, UK, 2002).
[2] Bar-Yam, Y. Dynamics of complex systems (Addison-
Wesley, 1997).
[3] Allen, B., Stacey, B. & Bar-Yam, Y. Multiscale infor-
mation theory and the marginal utility of information.
Entropy 19, 273 (2017).
[4] Bar-Yam, Y. Making things work: Solving complex prob-
lems in a complex world (Knowledge Press, 2004).
[5] Bouchaud, J.-P. Crises and collective socio-economic
phenomena: Simple models and challenges. Journal of
Statistical Physics 151, 567–606 (2013).
[6] Bar-Yam, Y. Complexity of military conflict: Multi-
scale complex systems analysis of littoral warfare. Re-
port to Chief of Naval Operations Strategic Studies Group
(2003).
[7] Bar-Yam, Y. Improving the effectiveness of health care
13
tems are those that, at each scale, match the complexity
of their environments.
When analyzing data or creating organizational struc-
tures, standard methods fail when they underestimate
the importance of interdependencies and the complexity
that arises from these interdependencies. To some extent,
these problems can be mitigated by matching the data
analysis or organizational structure to natural divisions
within the system. Since complex systems are those for
which behaviors occur over multiple scales, successful or-
ganizations and analyses for complex systems must also
be multi-scale in nature. However, even when armed with
all the proper information and tools, individual human
understanding of most complex systems will inevitably
fall short, with unpredictability being the best predic-
tion. To confront this reality, we must design systems
that, like evolution, are strengthened rather than weak-
ened by unpredictability. Such systems are flexible with
multiple processes occurring in parallel; these processes
may compete with one another within a multi-scale coop-
erative framework, such that effective practices are repli-
cated. Only these systems—that grow in complexity over
time from trial and error and the input of many—exhibit
the necessary complexity to solve problems that exceed
the limits of human comprehension.
ACKNOWLEDGMENTS
This material is based upon work supported by the Na-
tional Science Foundation Graduate Research Fellowship
under Grant No. 1122374 and the Hertz Foundation. We
thank Uyi Stewart for discussions that led to the writing
of this paper, Gwendolyn Towers for editing early drafts
of the manuscript, and Robi Bhattacharjee for helpful
discussions regarding the definitions of complexity and
scale.
and public health: A multiscale complex systems anal-
ysis. American Journal of Public Health 96, 459–466
(2006).
[8] Canzoneri, M. B., Cumby, R. E. & Diba, B. T. How do
monetary and fiscal policy interact in the European mon-
etary union? Tech. Rep., National Bureau of Economic
Research (2005).
[9] Semmler, W. & Zhang, W. Monetary and fiscal pol-
icy interactions in the euro area. Empirica 31, 205–227
(2004).
[10] Alessandrini, P. & Fratianni, M. U. In the absence of
fiscal union, the eurozone needs a more flexible monetary
policy. PSL Quarterly Review 68 (2015).
[11] Dan, H. The euro zone–between fiscal heterogeneity and
monetary unity. Transylvanian Review of Administrative
Sciences 43 E, 68–84 (2014).
[12] Drudi, F., Durré, A. & Mongelli, F. P. The interplay of
economic reforms and monetary policy: The case of the
eurozone. JCMS: Journal of Common Market Studies

50, 881–898 (2012).
[13] Davidson, A. W. & Bar-Yam, Y. Environmental com-
plexity: Information for human-environment well-being.
In Minai, A. A. & Bar-Yam, Y. (eds.) Unifying Themes
in Complex Systems, 157–168 (Springer, 2006).
[14] Nissani, M. Ten cheers for interdisciplinarity: The case
for interdisciplinary knowledge and research. The Social
Science Journal 34, 201–216 (1997).
[15] Brewer, G. D. The challenges of interdisciplinarity. Policy
Sciences 32, 327–337 (1999).
[16] Feller, I. New organizations, old cultures: Strategy and
implementation of interdisciplinary programs. Research
Evaluation 11, 109–116 (2002).
[17] Rhoten, D. & Parker, A. Risks and rewards of an in-
terdisciplinary research path. Science 306, 2046–2046
(2004).
[18] Lélé, S. & Norgaard, R. B. Practicing interdisciplinarity.
BioScience 55, 967–975 (2005).
[19] Jacobs, J. A. & Frickel, S. Interdisciplinarity: A criti-
cal assessment. Annual Review of Sociology 35, 43–65
(2009).
[20] Janis, I. L. Groupthink. IEEE Engineering Management
Review 36, 36 (2008).
[21] Trueman, B. Analyst forecasts and herding behavior.
The Review of Financial Studies 7, 97–124 (1994).
[22] Pan, W., Altshuler, Y. & Pentland, A. Decoding social
influence and the wisdom of the crowd in financial trading
network. In 2012 International Conference on Privacy,
Security, Risk and Trust and 2012 International Confer-
nece on Social Computing, 203–209 (IEEE, 2012).
[23] Bar-Yam, Y. From big data to important information.
Complexity 21, 73–98 (2016).
[24] Kadanoff, L. P. More is the same; phase transitions and
mean field theories. Journal of Statistical Physics 137,
777 (2009).
[25] Mora, T. & Bialek, W. Are biological systems poised at
criticality? Journal of Statistical Physics 144, 268–302
(2011).
[26] Davies, J. B., Lluberas, R. & Shorrocks, A. F. Estimating
the level and distribution of global wealth, 2000–2014.
Review of Income and Wealth 63, 731–759 (2017).
[27] Rauch, E. M. & Bar-Yam, Y. Long-range interactions
and evolutionary stability in a predator-prey system.
Physical Review E 73, 020903 (2006).
[28] Taleb, N. N. How much data do you need? an opera-
tional, pre-asymptotic metric for fat-tailedness. Interna-
tional Journal of Forecasting 35, 677 – 686 (2019).
[29] Kardar, M. Statistical Physics of Fields (Cambridge Uni-
14
versity Press, 2007).
[30] Lim, M., Metzler, R. & Bar-Yam, Y. Global pattern for-
mation and ethnic/cultural violence. Science 317, 1540–
1544 (2007).
[31] Rutherford, A. et al. Good fences: The importance of
setting boundaries for peaceful coexistence. PLOS One
9, e95660 (2014).
[32] Schelling, T. C. Dynamic models of segregation. Journal
of Mathematical Sociology 1, 143–186 (1971).
[33] Turing, A. M. Computing machinery and intelligence. In
Epstein, R., Roberts, G. & Beber, G. (eds.) Parsing the
Turing Test, 23–65 (Springer, 2009).
[34] McDonnell, M. D. & Ward, L. M. The benefits of noise in
neural systems: Bridging theory and experiment. Nature
Reviews Neuroscience 12, 415 (2011).
[35] Goldenfeld, N. & Woese, C. Life is physics: Evolution
as a collective phenomenon far from equilibrium. Annual
Review of Condensed Matter Physics 2, 375–399 (2011).
[36] Taleb, N. N. Antifragile: Things that gain from disorder
(Random House Incorporated, 2012).
[37] Olszak, T. et al. Microbial exposure during early life has
persistent effects on natural killer t cell function. Science
336, 489–493 (2012).
[38] Su, L. F., Kidd, B. A., Han, A., Kotzin, J. J. & Davis,
M. M. Virus-specific cd4+ memory-phenotype t cells are
abundant in unexposed adults. Immunity 38, 373–383
(2013).
[39] Seery, M. D. Resilience: A silver lining to experiencing
adverse life events? Current Directions in Psychological
Science 20, 390–394 (2011).
[40] Lukianoff, G. & Haidt, J. The coddling of the american
mind. The Atlantic 316, 42–52 (2015).
[41] Bar-Yam, Y. Complexity theory in applied policy world-
wide. Modeling complex systems for public policies. Alves
B, Furtado P, Sakowski M, Tóvolli M. Brası́lia: IPEA
(2015).
[42] Gard-Murray, A. S. & Bar-Yam, Y. Complexity and
the limits of revolution: What will happen to the arab
spring? In Fellman, P. V., Bar-Yam, Y. & Minai, A. A.
(eds.) Conflict and Complexity, 281–292 (Springer, 2015).
[43] Bar-Yam, Y. When systems engineering fails—toward
complex systems engineering. Exchange 16, 1990–1993
(2003).
[44] Werfel, J. & Bar-Yam, Y. The evolution of reproduc-
tive restraint through social communication. Proceedings
of the National Academy of Sciences 101, 11019–11024
(2004).
[45] Wilson, D. S. Does altruism exist?: Culture, genes, and
the welfare of others (Yale University Press, 2015).